Pre-trained models usually come with a pre-defined tokenization and little flexibility as to what subword tokens can be used in downstream tasks. This problem concerns especially multilingual NLP and low-resource languages, which are typically processed using cross-lingual transfer. In this paper, we aim to find out if the right granularity of tokenization is helpful for a text classification task, namely dialect classification. Aiming at generalizations beyond the studied cases, we look for the optimal granularity in four dialect datasets, two with relatively consistent writing (one Arabic and one Indo-Aryan set) and two with considerably inconsistent writing (one Arabic and one Swiss German set). To gain more control over subword tokenization and ensure direct comparability in the experimental settings, we train a CNN classifier from scratch comparing two subword tokenization methods (Unigram model and BPE). For reference, we compare the results obtained in our analysis to the state of the art achieved by fine-tuning pre-trained models. We show that models trained from scratch with an optimal tokenization level perform better than fine-tuned classifiers in the case of highly inconsistent writing. In the case of relatively consistent writing, fine-tuned models remain better regardless of the tokenization level.

We present STT4SG-350, a corpus of Swiss German speech, annotated with Standard German text at the sentence level. The data is collected using a web app in which the speakers are shown Standard German sentences, which they translate to Swiss German and record. We make the corpus publicly available. It contains 343 hours of speech from all dialect regions and is the largest public speech corpus for Swiss German to date. Application areas include automatic speech recognition (ASR), text-to-speech, dialect identification, and speaker recognition. Dialect information, age group, and gender of the 316 speakers are provided. Genders are equally represented and the corpus includes speakers of all ages. Roughly the same amount of speech is provided per dialect region, which makes the corpus ideally suited for experiments with speech technology for different dialects. We provide training, validation, and test splits of the data. The test set consists of the same spoken sentences for each dialect region and allows a fair evaluation of the quality of speech technologies in different dialects. We train an ASR model on the training set and achieve an average BLEU score of 74.7 on the test set. The model beats the best published BLEU scores on 2 other Swiss German ASR test sets, demonstrating the quality of the corpus.

We present the TeDDi sample, a diversity sample of text data for language comparison and multilingual Natural Language Processing. The TeDDi sample currently features 89 languages based on the typological diversity sample in the World Atlas of Language Structures. It consists of more than 20k texts and is accompanied by open-source corpus processing tools. The aim of TeDDi is to facilitate text-based quantitative analysis of linguistic diversity. We describe in detail the TeDDi sample, how it was created, data availability, and its added value through for NLP and linguistic research.

In this paper, we describe our systems submitted to the NADI Subtask 1: country-wise dialect classifications. We designed two types of solutions. The first type is convolutional neural network CNN) classifiers trained on subword segments of optimized lengths. The second type is fine-tuned classifiers with BERT-based language specific pre-trained models. To deal with the missing dialects in one of the test sets, we experimented with binary classifiers, analyzing the predicted probability distribution patterns and comparing them with the development set patterns. The better performing approach on the development set was fine-tuning language specific pre-trained model (best F-score 26.59%). On the test set, on the other hand, we obtained the best performance with the CNN model trained on subword tokens obtained with a Unigram model (the best F-score 26.12%). Re-training models on samples of training data simulating missing dialects gave the maximum performance on the test set version with a number of dialects lesser than the training set (F-score 16.44%)

Pre-trained multilingual models, such as mBERT, XLM-R and mT5, are used to improve the performance on various tasks in low-resource languages via cross-lingual transfer. In this framework, English is usually seen as the most natural choice for a transfer language (for fine-tuning or continued training of a multilingual pre-trained model), but it has been revealed recently that this is often not the best choice. The success of cross-lingual transfer seems to depend on some properties of languages, which are currently hard to explain. Successful transfer often happens between unrelated languages and it often cannot be explained by data-dependent factors. In this study, we show that languages written in non-Latin and non-alphabetic scripts (mostly Asian languages) are the best choices for improving performance on the task of Masked Language Modelling (MLM) in a diverse set of 30 low-resource languages and that the success of the transfer is well predicted by our novel measure of Subword Evenness (SuE). Transferring language models over the languages that score low on our measure results in the lowest average perplexity over target low-resource languages. Our correlation coefficients obtained with three different pre-trained multilingual models are consistently higher than all the other predictors, including text-based measures (type-token ratio, entropy) and linguistically motivated choice (genealogical and typological proximity).

Cross-lingual transfer of parsing models has been shown to work well for several closely-related languages, but predicting the success in other cases remains hard. Our study is a comprehensive analysis of the impact of linguistic distance on the transfer of UD parsers. As an alternative to syntactic typological distances extracted from URIEL, we propose three text-based feature spaces and show that they can be more precise predictors, especially on a more local scale, when only shorter distances are taken into account. Our analyses also reveal that the good coverage in typological databases is not among the factors that explain good transfer.

This paper deals with the problem of incre-mental dialect identification. Our goal is toreliably determine the dialect before the fullutterance is given as input. The major partof the previous research on dialect identification has been model-centric, focusing on performance. We address a new question: How much input is needed to identify a dialect? Ourapproach is a data-centric analysis that resultsin general criteria for finding the shortest inputneeded to make a plausible guess. Workingwith three sets of language dialects (Swiss German, Indo-Aryan and Arabic languages), weshow that it is possible to generalize across dialects and datasets with two input shorteningcriteria: model confidence and minimal inputlength (adjusted for the input type). The sourcecode for experimental analysis can be found atGithub.

Neural models for morphological inflection have recently attained very high results. However, their interpretation remains challenging. Towards this goal, we propose a simple linguistically-motivated variant to the encoder-decoder model with attention. In our model, character-level cross-attention mechanism is complemented with a self-attention module over substrings of the input. We design a novel approach for pattern extraction from attention weights to interpret what the model learn. We apply our methodology to analyze the model’s decisions on three typologically-different languages and find that a) our pattern extraction method applied to cross-attention weights uncovers variation in form of inflection morphemes, b) pattern extraction from self-attention shows triggers for such variation, c) both types of patterns are closely aligned with grammar inflection classes and class assignment criteria, for all three languages. Additionally, we find that the proposed encoder attention component leads to consistent performance improvements over a strong baseline.

The distributions of orthographic word types are very different across languages due to typological characteristics, different writing traditions and potentially other factors. The wide range of cross-linguistic diversity is still a major challenge for NLP and the study of language. We use BPE and information-theoretic measures to investigate if distributions become similar under specific levels of subword tokenization. We perform a cross-linguistic comparison, following incremental merges of BPE (we go from characters to words) for 47 diverse languages. We show that text entropy values (a feature of probability distributions) tend to converge at specific subword levels: relatively few BPE merges (around 350) lead to the most similar distributions across languages. Additionally, we analyze the interaction between subword and word-level distributions and show that our findings can be interpreted in light of the ongoing discussion regarding different types of morphological complexity.

We introduce a dictionary containing normalized forms of common words in various Swiss German dialects into High German. As Swiss German is, for now, a predominantly spoken language, there is a significant variation in the written forms, even between speakers of the same dialect. To alleviate the uncertainty associated with this diversity, we complement the pairs of Swiss German - High German words with the Swiss German phonetic transcriptions (SAMPA). This dictionary becomes thus the first resource to combine large-scale spontaneous translation with phonetic transcriptions. Moreover, we control for the regional distribution and insure the equal representation of the major Swiss dialects. The coupling of the phonetic and written Swiss German forms is powerful. We show that they are sufficient to train a Transformer-based phoneme to grapheme model that generates credible novel Swiss German writings. In addition, we show that the inverse mapping - from graphemes to phonemes - can be modeled with a transformer trained with the novel dictionary. This generation of pronunciations for previously unknown words is key in training extensible automated speech recognition (ASR) systems, which are key beneficiaries of this dictionary.

Strong regional variation, together with the lack of standard orthography, makes Swiss German automatic speech recognition (ASR) particularly difficult in a multi-dialectal setting. This paper focuses on one of the many challenges, namely, the choice of the output text to represent non-standardised Swiss German. We investigate two potential options: a) dialectal writing – approximate phonemic transcriptions that provide close correspondence between grapheme labels and the acoustic signal but are highly inconsistent and b) normalised writing – transcriptions resembling standard German that are relatively consistent but distant from the acoustic signal. To find out which writing facilitates Swiss German ASR, we build several systems using the Kaldi toolkit and a dataset covering 14 regional varieties. A formal comparison shows that the system trained on the normalised transcriptions achieves better results in word error rate (WER) (29.39%) but underperforms at the character level, suggesting dialectal transcriptions offer a viable solution for downstream applications where dialectal differences are important. To better assess word-level performance for dialectal transcriptions, we use a flexible WER measure (FlexWER). When evaluated with this metric, the system trained on dialectal transcriptions outperforms that trained on the normalised writing. Besides establishing a benchmark for Swiss German multi-dialectal ASR, our findings can be helpful in designing ASR systems for other languages without standard orthography.

In this paper, we present the findings of the Third VarDial Evaluation Campaign organized as part of the sixth edition of the workshop on Natural Language Processing (NLP) for Similar Languages, Varieties and Dialects (VarDial), co-located with NAACL 2019. This year, the campaign included five shared tasks, including one task re-run – German Dialect Identification (GDI) – and four new tasks – Cross-lingual Morphological Analysis (CMA), Discriminating between Mainland and Taiwan variation of Mandarin Chinese (DMT), Moldavian vs. Romanian Cross-dialect Topic identification (MRC), and Cuneiform Language Identification (CLI). A total of 22 teams submitted runs across the five shared tasks. After the end of the competition, we received 14 system description papers, which are published in the VarDial workshop proceedings and referred to in this report.

We present the results and the findings of the Second VarDial Evaluation Campaign on Natural Language Processing (NLP) for Similar Languages, Varieties and Dialects. The campaign was organized as part of the fifth edition of the VarDial workshop, collocated with COLING’2018. This year, the campaign included five shared tasks, including two task re-runs – Arabic Dialect Identification (ADI) and German Dialect Identification (GDI) –, and three new tasks – Morphosyntactic Tagging of Tweets (MTT), Discriminating between Dutch and Flemish in Subtitles (DFS), and Indo-Aryan Language Identification (ILI). A total of 24 teams submitted runs across the five shared tasks, and contributed 22 system description papers, which were included in the VarDial workshop proceedings and are referred to in this report.

Text normalization is the task of mapping non-canonical language, typical of speech transcription and computer-mediated communication, to a standardized writing. It is an up-stream task necessary to enable the subsequent direct employment of standard natural language processing tools and indispensable for languages such as Swiss German, with strong regional variation and no written standard. Text normalization has been addressed with a variety of methods, most successfully with character-level statistical machine translation (CSMT). In the meantime, machine translation has changed and the new methods, known as neural encoder-decoder (ED) models, resulted in remarkable improvements. Text normalization, however, has not yet followed. A number of neural methods have been tried, but CSMT remains the state-of-the-art. In this work, we normalize Swiss German WhatsApp messages using the ED framework. We exploit the flexibility of this framework, which allows us to learn from the same training data in different ways. In particular, we modify the decoding stage of a plain ED model to include target-side language models operating at different levels of granularity: characters and words. Our systematic comparison shows that our approach results in an improvement over the CSMT state-of-the-art.

Learning internal word structure has recently been recognized as an important step in various multilingual processing tasks and in theoretical language comparison. In this paper, we present a neural encoder-decoder model for learning canonical morphological segmentation. Our model combines character-level sequence-to-sequence transformation with a language model over canonical segments. We obtain up to 4% improvement over a strong character-level encoder-decoder baseline for three languages. Our model outperforms the previous state-of-the-art for two languages, while eliminating the need for external resources such as large dictionaries. Finally, by comparing the performance of encoder-decoder and classical statistical machine translation systems trained with and without corpus counts, we show that including corpus counts is beneficial to both approaches.

The paper documents the procedure of building a new Universal Dependencies (UDv2) treebank for Serbian starting from an existing Croatian UDv1 treebank and taking into account the other Slavic UD annotation guidelines. We describe the automatic and manual annotation procedures, discuss the annotation of Slavic-specific categories (case governing quantifiers, reflexive pronouns, question particles) and propose an approach to handling deverbal nouns in Slavic languages.

Language complexity is an intriguing phenomenon argued to play an important role in both language learning and processing. The need to compare languages with regard to their complexity resulted in a multitude of approaches and methods, ranging from accounts targeting specific structural features to global quantification of variation more generally. In this paper, we investigate the degree to which morphological complexity measures are mutually correlated in a sample of more than 500 languages of 101 language families. We use human expert judgements from the World Atlas of Language Structures (WALS), and compare them to four quantitative measures automatically calculated from language corpora. These consist of three previously defined corpus-derived measures, which are all monolingual, and one new measure based on automatic word-alignment across pairs of languages. We find strong correlations between all the measures, illustrating that both expert judgements and automated approaches converge to similar complexity ratings, and can be used interchangeably.

In this paper we present a newly developed tool that enables researchers interested in spatial variation of language to define a geographic perimeter of interest, collect data from the Twitter streaming API published in that perimeter, filter the obtained data by language and country, define and extract variables of interest and analyse the extracted variables by one spatial statistic and two spatial visualisations. We showcase the tool on the area and a selection of languages spoken in former Yugoslavia. By defining the perimeter, languages and a series of linguistic variables of interest we demonstrate the data collection, processing and analysis capabilities of the tool.

Swiss dialects of German are, unlike most dialects of well standardised languages, widely used in everyday communication. Despite this fact, automatic processing of Swiss German is still a considerable challenge due to the fact that it is mostly a spoken variety rarely recorded and that it is subject to considerable regional variation. This paper presents a freely available general-purpose corpus of spoken Swiss German suitable for linguistic research, but also for training automatic tools. The corpus is a result of a long design process, intensive manual work and specially adapted computational processing. We first describe how the documents were transcribed, segmented and aligned with the sound source, and how inconsistent transcriptions were unified through an additional normalisation layer. We then present a bootstrapping approach to automatic normalisation using different machine-translation-inspired methods. Furthermore, we evaluate the performance of part-of-speech taggers on our data and show how the same bootstrapping approach improves part-of-speech tagging by 10% over four rounds. Finally, we present the modalities of access of the corpus as well as the data format.

Verb aspect is a grammatical and lexical category that encodes temporal unfolding and duration of events described by verbs. It is a potentially interesting source of information for various computational tasks, but has so far not been studied in much depth from the perspective of automatic processing. Slavic languages are particularly interesting in this respect, as they encode aspect through complex and not entirely consistent lexical derivations involving prefixation and suffixation. Focusing on Croatian and Serbian, in this paper we propose a novel framework for automatic classification of their verb types into a number of fine-grained aspectual classes based on the observable morphology of verb forms. In addition, we provide a set of around 2000 verbs classified based on our framework. This set can be used for linguistic research as well as for testing automatic classification on a larger scale. With minor adjustments the approach is also applicable to other Slavic languages

We present a novel tool for morphological analysis of Serbian, which is a low-resource language with rich morphology. Our tool produces lemmatisation and morphological analysis reaching accuracy that is considerably higher compared to the existing alternative tools: 83.6% relative error reduction on lemmatisation and 8.1% relative error reduction on morphological analysis. The system is trained on a small manually annotated corpus with an approach based on Bidirectional Sequence Classification and Guided Learning techniques, which have recently been adapted with success to a broad set of NLP tagging tasks. In the system presented in this paper, this general approach to tagging is applied to the lemmatisation task for the first time thanks to our novel formulation of lemmatisation as a category tagging task. We show that learning lemmatisation rules from annotated corpus and integrating the context information in the process of morphological analysis provides a state-of-the-art performance despite the lack of resources. The proposed system can be used via a web GUI that deploys its best scoring configuration